How many ML tree should I build?

Question

I created a phylogenetic tree of 328 species using RAxML. I run the ML algorithm 20 times and I chose the best tree (the one with the highest likelihood).

The choice of 20 iterations was based on a tutorial that was carrying out the same analysis but on fewer species.

Now I have a doubt that my search is actually not optimal, and could have got better by increasing the number of iterations.

Do you know if there is a way to estimate the optimal number of iterations to find the best robust ML Tree?

Answer 1

Simply use the iterations option (e.g. -n 1000) of IQTREE. It's a lot quicker (several logs) and is strongly of identical performance to RAxML.

iqtree -s alignment.phy -n 1000 -m GTR+I+G -nt 4

It takes phylip, fasta, nexus ... and no-one questions 1000 iterations, so you're safe.

I personally I'd then go on to bootstrap the tree (IQTREE is extremely fast). If you were bootstrapping with iterations RAxML is unworkable, you'd need e.g. 10 iterations per bootstrap (1000 bootstraps) and would be time consuming. IQTREE could easily perform 10 iterations per bootstrap and would accommodate 100 iterations (1000 would be a bit excessive for bootstrapping).

RAxML is an old program now, IQTREE is its successor and massively refactored for maximum likelihood phylogeny.

---

If you don't understand the command line, drop a note and I'll explain the model it's using.